cloud-architecture-awslisted

# AWS Architecture Knowledge Service-level detail for an AWS-backed project. Pairs with Melvin's cloud-agnostic diagnostic checklist (traffic pattern, state location, SLAs, blast radius, cost explosion, coordination, limits, observability) — this pack supplies the AWS-specific answers for each. For concrete project topology, cost tiers, and stack context, read the architecture docs named in `CLAUDE.md`. ## Compute ### Lambda - **Cold starts** are the p99 killer. Node/Python ~100-400ms; JVM/.NET worse; VPC-attached used to add seconds (now ~sub-100ms with Hyperplane ENI, but still non-zero). A function called rarely is *always* cold. - **Mitigations:** provisioned concurrency (you pay for warm instances), keeping the bundle small (esbuild/tree-shake), avoiding heavy module-level init, SnapStart for JVM. Provisioned concurrency defeats the "scale to zero" cost story — only buy it where the latency SLA demands it. - **Concurrency limits:** default account limit is 1,000 concurrent executions (raise via quota request). Burst concurrency adds a few thousand instantly then scales +500/min. A traffic spike past burst headroom = throttles (429), not infinite scale. - **Reserved concurrency** carves a function's share out of the account pool AND caps it — useful to protect a downstream (e.g. a DB) from a stampede, but it can starve other functions. - **15-minute max** execution; 10GB memory ceiling (CPU scales with memory). Payload 6MB sync / 256KB async. `/tmp